AIDRC Seminar Series - Konstantin Mishchenko

Title:

Federated Learning for Heterogenous Data Using ProxSkip and Friends

Abstract:

In this talk, we will cover a number of algorithms for federated learning with heterogeneous data. The presentation will be centered around the ProxSkip algorithm and its extensions by various authors to other settings. In particular, we will start off by discussing the ideas behind ProxSkip and then proceed to its variants with quantization and client sampling. The presentation will be finished with open problems and possible research directions.

Bio:

Konstantin Mishchenko is a Research Scientist at Samsung in Cambridge, UK, working on optimization theory and federated learning. He did his PhD under the supervision of Peter Richtárik from 2017 to 2021 and did a postdoc in the group of Francis Bach at Inria Paris in 2021. Konstantin has been recognized as an outstanding reviewer for NeurIPS19, ICML20, AAAI20, ICLR21, ICML21, NeurIPS21, ICLR22, ICML22 and served as an Area Chair for ACML 2022. He was named a Rising Career in Data Science by the University of Chicago in 2021 and has published 11 conference papers at ICML, ICLR, NeurIPS, AISTATS, and UAI.